regression: * deps no longer parsed

[Manishearth]

On cargo 0.13.0-nightly (c205132 2016-08-09), deps like 0.2* no longer parse.

I'm not sure if these are valid, but they used to work, e.g on stable (cargo 0.11.0-nightly (259324c 2016-05-20))

Minimal example:

[package]

name = "foo"
version = "0.3.0"

[lib]
name = "foo"
path = "src/lib.rs"

[dependencies]
plugin = "0.2*"

works fine on stable, gives the following on nightly:

$ cargo build --verbose
error: failed to parse manifest at `/Users/manishearth/sand/bar/tempcrate/Cargo.toml`

Caused by:
  you have already provided an operation, such as =, ~, or ^; only use one

[steveklabnik]

They are not valid per semver, you need the .

[Manishearth]

Can we then phase them out slowly? This is a breaking change that affects people pinning to iron 3 (e.g. rust-playpen).

[steveklabnik]

It's going to be a giant pain for me.

We could always revert cargo back to a previous semver version as a quick fix.

[steveklabnik]

So, thinking more: I could backport a warning to the 2.x series of semver, we could make cargo use that instead. That probably wouldn't be too bad, though part of the reason why I re-wrote all that code was because it is a giant mess...

[Manishearth]

Yeah, I don't want to muck up semver's code too much. But I'm not sure how this can be done on the pure-cargo side either.

Perhaps a crater run would help us gauge the full impact and determine if this is necessary? @brson

[steveklabnik]

So, another thing:

I currently do a regression test across all of crates.io: https://github.com/steveklabnik/semver/blob/master/tests/regression.rs

But note that's just for the version, not for dependencies. it looks like that's supported https://frewsxcv.github.io/rust-crates-index/crates_index/struct.Dependency.html

So I could also figure it out that way, and probably should add that regardless.

[steveklabnik]

So, uh https://travis-ci.org/steveklabnik/semver/jobs/152783177

That is, this seemgly passed on semver master for every crate on crates.io.

@Manishearth I can't actually find a released iron that has this dep. v3.0: iron/iron@16c858e

Shows them actually fixing this. And it looks like 2.6 wasn't tagged?

[steveklabnik]

I used cargo-clone to grab iron 2.6, and it does have 0.2*

So, now I'm confused about my tests :(

[steveklabnik]

okay so https://travis-ci.org/steveklabnik/semver/jobs/152791793

testing crate iron at version: 0.2.6

testing dependency: ^0.1

testing dependency: ^0.2

testing dependency: ^0.7

testing dependency: ^0.3

testing dependency: 0.2.*

testing dependency: ^0.3

testing dependency: ^0.1

testing dependency: ^0.5

testing dependency: ^0.1

testing dependency: ^0.7

apparently, the crates-index crate is not returning the same contents as the actual cargo.toml.

Looking at https://github.com/rust-lang/crates.io-index/blob/master/ir/on/iron#L47

{"name":"iron","vers":"0.2.6","deps":[{"name":"modifier","req":"^0.1","features":[],"optional":false,"default_features":true,"target":null,"kind":"normal"},{"name":"num_cpus","req":"^0.2","features":[],"optional":false,"default_features":true,"target":null,"kind":"normal"},{"name":"conduit-mime-types","req":"^0.7","features":[],"optional":false,"default_features":true,"target":null,"kind":"normal"},{"name":"log","req":"^0.3","features":[],"optional":false,"default_features":true,"target":null,"kind":"normal"},{"name":"plugin","req":"0.2.*","features":[],"optional":false,"default_features":true,"target":null,"kind":"normal"},{"name":"typemap","req":"^0.3","features":[],"optional":false,"default_features":true,"target":null,"kind":"normal"},{"name":"error","req":"^0.1","features":[],"optional":false,"default_features":true,"target":null,"kind":"normal"},{"name":"url","req":"^0.5","features":[],"optional":false,"default_features":true,"target":null,"kind":"normal"},{"name":"lazy_static","req":"^0.1","features":[],"optional":false,"default_features":true,"target":null,"kind":"normal"},{"name":"hyper","req":"^0.7","features":[],"optional":false,"default_features":false,"target":null,"kind":"normal"},{"name":"time","req":"*","features":[],"optional":false,"default_features":true,"target":null,"kind":"dev"}],"cksum":"6d02706746b2d95c50f463fc619de64efe10682adfa17fa18dc95e0a92b10237","features":{"default":["ssl"],"ssl":["hyper/ssl"]},"yanked":false}

This also has 0.2.*.

@alexcrichton do you have any idea what's going on here?

[steveklabnik]

From IRC:

20:43 <~acrichto> steveklabnik: I think we round trip
20:43 <~acrichto> in the sense of we parse the requirement
20:43 <~acrichto> and then when we emit the json we emit that requirement to a string
20:43 < steveklabnik> would that be in cargo or in crates.io?
20:45 <~acrichto> steveklabnik: oh probably both actually
20:45 <~acrichto> as in, Cargo's sending requests to crates.io
20:45 <~acrichto> but crates.io is then re-parsing and re-encoding

So yeah, @brson a crater run to check this out might be a good idea. idk. uuuugh

[steveklabnik]

That said, this change has been in nightly for a while, and we also haven't seen many people file issues about breakage. So maybe things are okay? I dunno 😦

[alexcrichton]

Hm ok, so I think we'll want to revert this for now and go back to the old semver version. @steveklabnik could you send a PR for that? Otherwise I can do so later tonight.

A similar situation here is that I fixed a parsing bug in TOML but Cargo's still issuing warnings about it. From time to time I still see that url's 0.5.* manifest doesn't parse correctly, so it'll take awhile to flesh all this out.

[steveklabnik]

@alexcrichton

It doesn't look like you ended up doing it, so let me see if I can cut a new version of the older semver with some kind of warning about this. If I can't get it done quickly, I'll just send it for the exiting version.

so it'll take awhile to flesh all this out.

Yeah, we should talk about strategies here.